Nonlinear adaptive image filtering based on inhomogeneous diffusion and differential geometry

摘要: The inadequacy of the classic linear approach to edge detection and scale space filtering lies in spatial averaging Laplacian. Laplacian is divergence gradient thus both magnitude direction. characterizes edges this must not be averaged if image structure preserved. We introduce a new nonlinear theory that only averages This keeps lines intact as their direction nondivergent. Noise does have nondivergent consistency its divergent averaged. Higher order structures such corners are singular points or inflection also Corners intersection (or smooth curves small direction) limited. provides better compromise between noise removal preservation structure. Experiments verify demonstrate adequacy presented.